67
Chapter 3 – Full and Partial Wh-Movement in Kashmiri
1. Introduction
Kashmiri exhibits both full and partial wh-movement as question formation strategies
in sentences with multiple clauses.
(1) tse k´m' chu-y ba:sa:n ki mohn-as dits kita:b
you who aux think that Mohan-dat gave book
‘Who do you think gave Mohan the book?’ (PK 9/21/04)
(2) tse k'a: chu-y ba:sa:n ki mohn-as k´m' dits kita:b
you expl aux think that Mohan-dat who gave book
‘Who do you think gave Mohan the book?’ (Wali and Koul: 18)
The question word in the subordinate clause receives a matrix scope interpretation in
both (1) and (2). In (1) this interpretation is achieved by movement of the question
word k´m' ' 'who' into the matrix clause. In (2) this is achieved via the presence of a
minimal question word k'a ('what') in the scope position in the matrix clause, while
the contentful question word k´m' remains in the subordinate clause.
Previous approaches to these constructions generally fall into two classes. So-
called ‘direct dependency’ analyses (McDaniel 1989; Rizzi 1992; Mahajan 1990;
inter alia) contend that a direct syntactic connection is formed between the
meaningless question word and the contentful question word, mediated by chains and
conditions on chain formation. The second class of approaches, called ‘indirect
68
dependency’ accounts (Kiss 1987; Dayal 1994, 1996; inter alia), denies that any such
syntactic connection exists. Instead they claim that the meaningless question word is
coindexed with or replaced by the clause containing the contentful question word at
the level of Logical Form (LF), and this is how the correct interpretation is achieved.
The account proposed here falls squarely into neither of these classes. It does
build on important properties of each, maintaining that the meaningless question word
is base-generated in a clause-internal (not clause-peripheral) position as in the indirect
dependency approaches, but also that the role played by the meaningless question
word itself is entirely a syntactic one, as in direct dependency analyses. However, the
essential point of contrast with these two camps is the claim developed in this chapter
that there is no connection at all between the meaningless question word and the
contentful question word in a lower clause, whether syntactic or at some level of
interpretation. Instead, the role of the meaningless question word in the A-bar system
will be analyzed as comparable to that of nominal expletives in the A-system
(following in the spirit of recent approaches by Simpson (2000) and Fanselow and
Mahajan (2000)).
The issues ultimately at stake here are large – in particular the question of
whether the syntax of the left periphery (the A-bar system) is governed by the same
organizing principles as clause-internal syntax (the A-system). If the proposals
developed here are on the right track, then the two systems emerge as being
completely parallel – in a way that does not emerge so clearly from the more closely
studied languages.
69
Section 2 of this chapter details important features of Kashmiri A-bar syntax,
particularly with respect to the way in which questions are formed. This section relies
on the account of the left periphery of the Kashmiri clause proposed in Chapter 2
above, and establishes a working view of the internal phrase structure of the Kashmiri
clause. Section 3 provides an account of wh-movement and wh-expletive
constructions in Kashmiri and compares the reach of this account with previous
approaches to wh-expletive phenomena. In Section 4, I discuss two remaining
questions, one internal to Kashmiri and one crosslinguistic in nature. Section 5
concludes the chapter.
2. Kashmiri Question Formation and the Structure of the Clause
2.1 The Kashmiri Question
Recall that Kashmiri is unusual among the Indic languages in exhibiting the
verb-second (V2) property, more familiar from Germanic and the older Romance
languages.
The unmarked word order of a Kashmiri tensed root clause is:
(3) subject-finite verb-indirect object-direct object
The finite verb must be the second constituent in the clause, but any of the arguments
(or other constituents) may appear first. The order of the postverbal elements is also
fairly free, though the subject must immediately follow the second-position verb if
the sentence is not subject-initial.
70
In constituent questions, the question word must appear before the verb in
addition to some other constituent. Only if the question word is the subject, may it
naturally appear alone in sentence-initial position.
(4) a. rajan kemis he:v nev kita:b?
Raj whom showed new book
‘To whom did Raj show his new book?’ (Wali and Koul: 12)
b. kem' he:v shi:las nev kita:b ra:th
who showed Sheila new book yesterday
‘Who showed a new book to Sheila yesterday?’ (Wali and Koul: 12)
Kashmiri has two strategies for forming constituent questions with more than one wh-
phrase. In the first, all wh-phrases are moved to the preverbal position. In the second,
only one wh-phrase is moved, and the remaining wh-phrases are found in-situ within
the clause.
(5) k´m' k´mis dits kita:b?
who whom gave book
‘Who gave the book to whom?’ (Wali and Koul: 26)
(6) k´m' dits k´mis kita:b
who gave whom book
‘Who gave the book to whom?’ (Wali and Koul: 27)
Subordinate clauses are identical to matrix clauses in their word order, except
that they are optionally preceded by the particle ki. This particle is not counted in
71
determining verb-second position, and will be considered here to be inserted as the
marker of the CP phase boundary, as discussed in Chapter 2. A typical indirect
question is in (7).
(7) mi:ra:yi cha pata: ki k´mis dits mohnan kita:b
Mira aux know that who gave Mohan book
‘Mira knows who gave Mohan the book.’ . (Wali 2002)
Multiple wh-phrases can appear in indirect questions as well, just as they can in
matrix clauses.
(8) Me chu nI pata: ki kus k´mis o:s me:lIni gatsha:n
I aux-not know that who whom bring meet going
‘I don’t know who was going to meet whom.’ (Wali and Koul: 27)
The focus of this chapter shall be constructions that permit matrix scope
interpretations of question words originating in subordinate clauses. Bridge verbs
permit just such a construction, in which an invariant wh-expletive k'a, appears in the
pre-verbal position in the matrix clause. We will call this the wh-expletive or partial
movement construction throughout. The specific behavior of non-bridge verbs with
respect to these constructions will be explored further below.
(12) tse k'a: chu-y ba:sa:n ki mohn-as kem' dits kita:b
you what aux think that Mohan who gave book
‘Who do you think gave Mohan the book?’
Direct questions formed by full wh-extraction from the clausal complements of these
72
non-factive verbs are also possible, as in (13).
(13) a. tse k'a:zi chuy ba:sa:n ki ra:jan a:si ditsmits; mohnas kita:b _
you why aux think that Raj aux gave Mohan book
‘Why do you think that Raj gave the book to Mohan?’
(Wali and Koul: 19)
b. mi:ra: k'a: chi yatsha:n ki su gotsh anun
Mira what aux want that he should bring
‘What does Mira want that he should bring?’ (Wali 2002)
2.2 Assumptions about the Structure of the Kashmiri Clause
The facts of Kashmiri could be construed to support an approach in which the
unmarked word order in Kashmiri [subject-verb-object] is derived by movement of
the verb to T and the subject to the specifier of TP. Bhatt (1999) reports that the
preverbal position is generally the focus position, and that elements in this position
always bear focal stress, with the important exception of subjects.1 This would
indicate that subjects may not necessarily be in the specifier of some head containing
focus features, but can instead be in the specifier of a head unmarked with respect to
focus. Putting these observations together, we are led to assume that in subject-initial
declarative clauses, the finite verb raises only as far as T. Unfocused subjects then
1 Bhatt also mentions that temporal adverbs do not appear stressed in this position. Why this is so is
beyond the scope of this discussion.
73
appear in Spec, TP. In the case of structures with focus and topic constituents, a C
bearing these features attracts the finite verbs to raise further.
This split view of verb-second is similar to that offered in Zwart (1997) (and
earlier in Travis 1991). According to Zwart, in Dutch (as in other Germanic
languages) subject-initial main clauses do not involve movement to CP, as has often
been assumed. Zwart also offers evidence against generalized V-to-C movement in
Dutch, demonstrating that there is no clear motivation for assuming that the verb has
always raised to C. From these considerations, Zwart concludes that the subject and
verb in subject-initial main clauses in Dutch are not located in CP, but instead in
AgrS, the highest head of the inflectional layer. He asserts that the only way a subject
can move into the specifier of CP is if it is attracted by some feature of CP (i.e. a wh-
feature) beyond that which involves the normal interaction with T. Only initial
constituents that are wh-words or non-subjects are analyzed by Zwart as raising to a
CP or higher phrase. It seems that this approach can also be applied to the verb-
second facts of Kashmiri, as we have seen above.
The proposal I have developed extensively in Chapter 2 and the suggestions I
have made here for verb-second in Kashmiri differ significantly from the account
offered in Bhatt (1999). Bhatt claims that verb-second crosslinguistically is the result
of movement of the verb and some sentential constituent to a functional projection
MoodP (MP). MP is a component of an articulated CP – a phrase that he asserts is
universal across languages whether or not they possess explicit morphological mood
markers (Kashmiri does not). Bhatt directly addresses Zwart's (1997) claims
74
concerning Dutch V2, arguing against the approach primarily on theory-internal
grounds. It seems that most or all of his concerns disappear when the proposal is
updated along the lines of the Minimalist Program as it is presented in Chomsky
(2000). The empirical argument he makes against Zwart's proposal involves
subordinate clauses and complementizers in Swedish, and I will not address these
issues here. The only mention of Kashmiri in Bhatt's discussion of Zwart's proposal is
that under the view that only in subject-initial sentences is the verb not in CP, it
would be difficult to explain that sentence-initial temporal adverbs in Kashmiri do not
receive sentential (i.e. focal) stress. If there are other reasons why such adverbs may
not receive sentential focus intonation, we can move forward in explaining why
subjects behave differently from all other sentence-initial constituents – because (as
in many languages) they may remain in Spec, TP. These issues do not bear on my
central goals in this chapter, and so I will proceed on the assumption that something
like Zwart’s approach is correct for Kashmiri.
This chapter and this dissertation are primarily concerned with the left
periphery of Kashmiri, and in particular the CP layer. However it is useful at this
point to establish what our basic assumptions will be about the internal structure of
the Kashmiri clause.
Bhatt (1999) claims that the placement of the verb in Kashmiri indicates that
lexical projections in Kashmiri are head-final, and functional projections are head-
initial. Overall, Kashmiri exhibits the properties of a head-final language. For
instance, adpositions appear following their complement.
75
(14) … ba:gas manz
… garden in
‘… in the garden’ (Wali and Koul: 45)
The verb appears in clause final position in non-finite and relative clauses, and when
the tensed auxiliary appears in second position the main verb is still clause final (as in
(16)).
(15) [yi ba khyva:n chus] su chuyi-aa tse khosh kara:n
what I eat aux that aux-Q you like do
‘Do you like what I eat?’ (Bhatt 1999)
(16) ba chu-s lark-as kita:b diva:n
I aux-msg boy-dat book give
‘I give a book to the boy.’ (Bhatt 1999)
On the other hand, the inflected verb appears in second position in all tensed clauses,
as is exemplified in (14) and (16) and discussed extensively in Chapter 2. This means
that the functional projection at the edge of the clause (CP) must have its head on the
left to arrive at the grammatical verb-second word ordering (otherwise, raising to C
would be string-vacuous). Following Bhatt and based on these facts, I will assume the
following structure in (17) for basic Kashmiri clauses.
76
(17) CP C TP T vP Subject
v VP Object V Verb The external argument, complement to the verb, and the verb in (14) are in the
positions into which they first merge in the structure. In the course of the derivation
of a simple subject-initial declarative, the external argument would raise to Spec, TP,
and the verb to T. If any other constituent is initial, it will merge into Spec, CP and
the verb will raise to C. In addition to the above, I will assume, following Bhatt
(1999) and Bayer’s (1996) claims for Bengali and Hindi-Urdu, that subordinate
clauses in Kashmiri are complements of the verb that are exceptionally taken on the
right. This will be discussed further for the case of Hindi-Urdu in Chapter 4.
3. Analyzing Full and Partial Wh-Movement in Kashmiri
With an understanding of Kashmiri phrase structure in place, based on the
account presented in Chapter 2 and above in this chapter, we can now take a closer
look at wh-movement in Kashmiri, in particular full and partial movement from
subordinate clauses. In section 3.1 I will present in detail a new approach to A-bar
77
movement that accounts for the facts in Kashmiri. In 3.2, I contrast this account with
previous approaches to partial movement constructions, indicating theoretical and
empirical differences. Finally, section 3.3 suggests how the interpretation of partial
movement constructions might proceed under the syntactic account proposed here.
3.1 A New Account of A-bar movement
Recall the basic data under discussion. Kashmiri exhibits both full and partial wh-
movement (wh-expletive constructions) as question formation strategies in sentences
with multiple clauses.
(1) tse k´m' chu-y ba:sa:n ki mohn-as dits kita:b
you who aux think that Mohan gave book
‘Who do you think gave Mohan the book?’
(2) tse k'a: chu-y ba:sa:n ki mohn-as k´m' dits kita:b
you expl aux think that Mohan who gave book
‘Who do you think gave Mohan the book?’
The current theoretical framework provides a feature-based understanding of A-
movement. It is worthwhile to consider approaching A-bar movement in the same way.
Following the spirit of the recent work of Simpson (2000) and Fanselow and Mahajan
(2000), I argue that the analysis of these two systems can be unified, using Kashmiri as
a test case for this hypothesis. I will show that both full extraction and partial wh-
movement of question words in Kashmiri can be analyzed using a system of
78
interpretable and uninterpretable features in a manner similar to the approach to the
A-system.
The distinction between full extraction from subordinate clauses and partial wh-
movement in Kashmiri can be analyzed as the distinction between the operations
Move and static Agree to satisfy uninterpretable features. In the feature-based
approach to A-movement, a nominal enters into an agreement relation with a higher
accessible head. This Agree operation is simply a mutual exchange of information
between a head and a nominal bearing the relevant features – an exchange of
information which takes place in a particular structural configuration as follows:
(18) H ..... α .... β
a. H commands α, which in turn commands β
b. α is 'closer to' H than β
'Closer to' is defined in terms of asymmetric c-command. That is, α is 'closer to' H
than β iff H commands α, α commands β, and β does not command α (α commands
β iff β is contained within the sister of α). (Adger 2003).
In this configuration, it will be possible for the features of the head (the Probe)
and those of the nominal (the Goal) to mutually value one another. If the required
relation is not established, features remain unvalued, and the derivation will not result
in a well formed syntactic object. Once all of the features of an element are valued,
the element is inactive, and its participation in head-nominal interactions will be
limited.
79
Within the A-system, a nominal may Agree with a higher head, and may also
raise if the probing head has the EPP property. This composite operation is called
Move, and will be defined here as Agree + pied-piping + re-Merge.2 If the EPP is not
present, the uninterpretable features of the Probe and Goal may of course be valued
by static Agree over a distance, as described above (Chomsky 2000). I also assume
(following Chomsky 2004) that all and only uninterpretable features have unvalued
instances when they enter the derivation.
The operations described above are constrained by locality considerations.
Agree (and hence the composite operation built upon it, Move) can only take place
within a phase. A phase is the unit in which derivations proceed. A Probe can only
interact with Goals within its own phase or on the edge of the previous phase.
Elements that are not within the current phase (that are contained in previously
constructed phases) are not available. For the purposes of this chapter, I will focus on
the CP phase only, though Chapter 4 will introduce the relevance of the v-phase for
wh-movement (Rackowski and Richards 2005). Chapter 2 has set the stage for our
understanding of the rich left periphery of the clause as a single C-head with multiple
specifiers, and this understanding will become crucial here.
In this chapter I will claim that the A-bar system of Kashmiri functions
identically to the A system in these respects. Heads and wh-phrases will possess
interpretable and uninterpretable features. If a higher probe possesses the EPP in
2 The first merge is when element is merged into the 'workspace' from the numeration.
80
addition to other features, an accessible wh-phrase will undergo Move. Alternatively,
if the wh-expletive k'a is in the numeration, the merging of k'a can satisfy the EPP on
the Probe, much like an expletive in the A-system. If this occurs, the uninterpretable
features on the A-bar probe will be valued by interacting with an accessible wh-
phrase via static Agree over some distance.
There are three features at work controlling movement and agreement in the
A-bar system: the EPP (common to the A and A-bar systems), the [Q] feature, and the
[wh] feature. The [wh] feature is interpretable on wh-phrases and uninterpretable on
all heads, activating probes that interact with wh-phrases. The interpretable feature on
wh-phrases is its "wh-hood"; that which triggers the interpretation of the wh-phrase as
a Reinhart (1998)-style choice function variable. The feature [Q] is uninterpretable on
the wh-phrase but present and interpretable on the highest head in an A-bar
movement sequence. This interpretable [Q] feature marks the position at which a wh-
phrase will be interpreted. Like the category-defining features, it allows the hosting
head to enter into selectional relations, and is interpreted as an unselective binder of
(choice) function variables (Reinhart 1998). The role of these interpretable features
will be further explored in section 3.3.
In a Kashmiri subordinate clause, a wh-phrase must always move to the
specifier of the C-head containing the second-position verb. Wh-in situ is impossible
except in instances of multiple wh-phrases (at least one wh-phrase must raise). This
means that the C-head which contains the second position verb will necessarily
possess an uninterpretable [wh] feature and the EPP in all interrogative clauses, both
81
matrix and subordinate. It is the uninterpretable [wh] feature that makes the C an
active Probe, and it is the EPP which requires that the wh-phrase in its domain not
only Agree but also Move into its specifier. Within an interrogative sentence
comprised of a single clause, as in (19), the C-head will also possess the interpretable
[Q] feature, as in (20).
(19) rajan k´mis he:v nev kita:b?
Raj whom showed new book
‘To whom did Raj show his new book?’
(20) C
[Q-i]
[wh-u]
EPP
The presence of the interpretable [Q] feature signals the position at which the binder
of the wh-phrase variable will be introduced. This feature will value the
uninterpretable [Q] feature on the wh-phrase, and will both arrest the movement of
the wh-phrase and allow the sentence to be a well-formed syntactic object (with no
unvalued uninterpretable features). In addition, the scope of the wh-phrase is
determined at the position of the interpretable [Q] feature, in the sense that this is the
position from which the wh-indefinite is unselectively bound.
The goal of this section is ultimately to describe extraction and partial wh--
movement in Kashmiri subordinate clauses, so let us now consider a clause that is
82
embedded. In this scenario, the wh-phrase may not remain in the lower clause, but
instead must raise all the way to matrix scope position, as in (1), repeated here.
(1) tse k´m' chu-y ba:sa:n ki mohn-as dits kita:b
you who aux think that Mohan gave book
‘Who do you think gave Mohan the book?’
In a sentence like (1), the subordinate C-head will lack an interpretable [Q] feature
(the embedding verb basa:n does not select a question). A wh-phrase that has raised
to the specifier of this CP will still have uninterpretable features that require valuing
because it bears a [Q] feature which is not interpretable and which has no matching
feature on the embedded C head. The wh-phrase in the specifier of the subordinate
CP will be in the same phase as the matrix C-head. The matrix C-head will have an
uninterpretable [wh] feature, the EPP, and the interpretable [Q] feature, just as in
(20). As a Probe, it will find the wh-phrase in the specifier of the subordinate CP and
will enter into an Agree relation and Move with this wh-phrase and attract it to its
specifier. The wh-phrase will raise to the specifier of the matrix CP, and the result
will be full extraction. The particle ki is inserted post-syntactically, as discussed in
Chapter 2 and indicated by an arrow. This process and the features involved are
diagrammed in (21).
83
(21) ' CP tse ‘you k´m C TP 'who' wh-u ki Q-i … CP EPP chu-y Mohan-as ‘aux’ k´m’ C TP wh-u EPP … dits ‘gave’ (21) represents the extraction of the wh-phrase and its passage through the specifier
of the embedded CP into the matrix CP. At this point all uninterpretable features on
the wh-phrase are valued and it is rendered inactive, or frozen, in its scope position.
An obvious question arises at this point. If static Agree is one of the
operations available to the derivation, why can't the uninterpretable feature of the wh-
phrase be valued by the matrix C-head while it remains in the specifier of subordinate
CP? In this case, if no wh-expletive happens to be in the numeration and if the EPP
on the matrix C-head must be satisfied by a [wh]-bearing element, the EPP would not
be satisfied, and the derivation would therefore crash.
Let us now turn to the other strategy by which wh-phrases originating in
subordinate clauses take matrix scope in Kashmiri: the partial wh-movement or wh-
expletive construction as in (2).
84
(2) tse k'a: chu-y ba:sa:n ki mohn-as k´m' dits kita:b
you expl aux think that Mohan who gave book
‘Who do you think gave Mohan the book?’
In a partial movement construction, the subordinate C-head will once again have only
an uninterpreatable [wh] feature and the [EPP]. The numeration happens to contain a
wh-expletive k'a, which can be merged to satisfy the EPP on the matrix C Probe. This
expletive differs from a full wh-phrase in its feature content. As an expletive, it
consists entirely of uninterpretable features, and contributes to the syntactic
computation only an (uninterpretable) interrogative feature (written Q-u here). This
understanding of the wh-expletive has two consequences: (i) the expletive may
appear only in questions (ii) since it lacks a wh-feature altogether, it cannot render
inactive the probe in whose specifier it is merged. As a result, the matrix C-head will
still need to value its uninterpretable [wh] feature, and will therefore probe its
domain. It finds the wh-phrase in the specifier of the subordinate CP, and will enter
into static Agree with it. In this way, all uninterpretable features are valued. This
theoretical view takes wh-expletives to be the A-bar counterparts to expletives of the
A-system. That is, they do not contribute any interpretable features to the derivation,
but have only uninterpretable features. This means that they can only serve to satisfy
the EPP and permit the head's features to be valued statically by some other
accessible element (Simpson 2000). This process is diagramed in (22).
85
(22) ' CP tse ‘you’ k’a C TP 'expl' wh-u ki Q-i … CP EPP chu-y Mohan-as ‘aux’ k´m’ C TP ‘who’ wh-u EPP … dits ‘gave’
The analysis presented here accounts for both full wh-movement in Kashmiri
as in (1) and partial wh-movement as in (2), and provides an understanding of how C-
heads and wh-phrases interact in the course of forming long-distance wh-
dependencies. I have proposed that the two distinct strategies for construing matrix
scope for embedded wh-phrases are driven by identical mechanisms – that is, the
features involved in the derivation of each strategy are precisely the same. The wh-
movement and wh-expletive constructions, in this view, differ only in their
numerations. If a wh-expletive happens to be present in the numeration, it will be
merged into the matrix Spec, CP, allowing the features of the lower wh-phrase to be
valued via static agreement over a distance. If the expletive is not present, the wh-
phrase will itself raise to value the features on the matrix C-head. Either way, it will
be the wh-phrase from the embedded clause that will value these features. A number
86
of questions concerning the properties and distribution of wh-expletives still remain
to be answered, however, and these are addressed in the following sections.
3.2 Restrictions on Wh-expletives
Consider the ungrammaticality of the Kashmiri clause in (23a), or its equivalent in
German, (23b).
(23) a. *rajan k'a: he:v k´mis nev kita:b?
Raj expl showed who new book
Intended: ‘To whom did Raj show his new book?’
b. *Was glaubst du was?
expl believe you what?
Intended: ‘What do you believe?’
In such cases, a wh-expletive is in the same clause as the full wh-phrase whose
position of interpretation it is meant to indicate. If, as we have explained above, an
expletive can be merged into Spec, CP to satisfy the EPP and the uninterpretable [wh]
feature on the C-head can be valued via static Agree with a wh-phrase in its phase,
there should be no problem with (23a-b). It seems that wh-expletives, unlike DP-
expletives, are constrained by a kind of anti-locality. Compare the ungrammaticality
of the wh-expletive constructions in (23) with the English DP-expletive there, in (24),
which can appear in the same clause as its DP associate.
87
(24) There are three unicorns in the garden.
Simpson (2000), along with Horvath (1997) and Fanselow and Mahajan (2000),
claims that the anti-locality property of wh-expletives can actually be reduced to a
question of case. If the wh-expletive in fact needs case just like any other wh-phrase,
it must actually be merged into a case position in a sentence like (23a). However, this
is not possible, because the full wh-phrase has occupied the relevant position and
received this case.3 The wh-expletive’s need for case will block instances like (23a),
in which the case is instead being assigned to the full wh-phrase k´mis ‘who’, but
permit (2) in which there is no competitor for case in the matrix clause.
(2) tse k'a: chu-y ba:sa:n ki mohn-as k´m' dits kita:b
you expl aux think that Mohan who gave book
‘Who do you think gave Mohan the book?’
Further empirical support for this view comes from Kashmiri and the related
language Hindi-Urdu. In addition to k'a:, Kashmiri has a pleonastic element yi that
can be optionally inserted into a matrix clause.
(25) bi o:sus yi za:na:n ki seli:m gav ra:th rajas sit [Kashmiri]
I aux this know that Selim went yesterday Raj with
‘I knew that Selim went with Raj yesterday.’ (Wali and Koul: 48)
A similar construction exists in Hindi-Urdu, as shown in (26).
3 Note that the need for DP case would not alter the status of k’a: or was as a wh-expletive. It is still defective in the sense that it cannot value the uninterpretable features on the C-head to whose specifier it raises. This will force the uninterpretable [wh] feature on the C-head to seek another wh-phrase with
88
(26) Miriam yeh jantaa hai ki Haroun kis-se baat karegii. [Hindi-Urdu]
Miriam this know aux that Haroun who-with talk aux
‘Miriam knows who Haroun will talk to.’
In both Hindi-Urdu and Kashmiri the expletive object (yi or yeh) cannot coexist with
a wh-expletive, suggesting that they occupy the same case position in the clause.
(27) *Sita-ne yeh kyaa socaa ki ravii-ne kis-ko dekhaa? [Hindi-Urdu]
Sita this expl thought that ravi who saw
Intended: ‘Who did Sita think that Ravi saw?’
Now let us turn to the position at which the wh-expletive is first introduced
in the clause. Simpson (2000) claims that wh-expletives are base generated in the
specifier of the agreement projection AgrO, and when other DPs are present which
need to check case in this position, it will not be possible to generate the wh-
expletive. This would rule out examples like (23) and (27), in which another wh-DP
or the expletive object yeh must occupy this position. Updating this view to reflect the
framework of this chapter, we need to ensure that the wh-expletive is base generated
in a position at which it can have accusative case valued; that is, within the c-
command domain of the accusative case licenser transitive v. Further, we know that
the wh-expletive is not a semantic argument of any predicate (having no interpretable
features) and cannot be introduced by semantic selection. For this reason, the wh-
expletive must be merged into the specifier of a head that has the EPP property. The
which to interact. In addition, case features are uninterpretable, and so that part of the understanding of wh-expletives also remains constant.
89
EPP is a quasi-selectional feature which causes the head to seek an additional
specifier beyond those it needs to satisfy its core selectional requirements. Aspect is a
functional head in the c-command domain of v on which it would be reasonable to
posit the EPP property (Travis 1991, to appear). AspectP is the projection claimed to
introduce aspect morphology, such as the perfective suffix -mut on the Kashmiri past
verbal stem (Travis 1991, Bhatt 1999). The wh-expletive introduced into the specifier
of AspP has three features: an uninterpretable D feature, an uninterpretable case
feature (accusative), and the uninterpretable Q feature. Note that it has no
interpretable features at all. When the transitive v is introduced, it will interact with
the wh-expletive and the expletive’s uninterpretable accusative case feature and [D]
feature will be valued. By the time the C-head is introduced, the wh-expletive has
only an unvalued uninterpretable [Q] feature remaining, and will interact with the C-
head in the way described above.
In this view, examples like (23) and (27) are impossible because only one
goal can interact with the v head and have its uninterpretable case feature valued. If
there is more than one potential goal, such as an additional wh-phrase or a clausal
expletive, the uninterpretable case feature on one or the other will go unvalued, and
the derivation will fail to converge.
Observations about Hungarian provided by Horvath (1997) offer additional
empirical evidence for this proposal. Hungarian is relevant because it has a partial
wh-movement construction whose properties closely parallel those of Kashmiri (and
Hindi-Urdu). The wh-expletive in the matrix clause actually exhibits case
90
morphology appropriate to its role in relation to the matrix verb, and the wh-phrase in
the subordinate clause is assigned a separate case, determined by properties of the
embedded clause, as expected.
(28) Mit mondtal hogy kinek vett Janos szinhazjegyet?
expl-acc said that who-dat bought John theatere ticket-acc
‘Who did you say John bought a theatre ticket for?’
Horvath originally intended this data to argue against the so-called 'direct
dependency', or chain-based approaches to partial wh-movement, claiming that the
chain itself could not be assigned different cases at its head and tail. Note that under
the analysis outlined here, (28) would not pose such a problem, because the wh-
phrase and wh-expletive are not syntactically connected, and can therefore be
assigned different cases as necessary.
The ramifications of the claim that wh-expletives (at least in some languages)
need case go beyond explaining the ungrammaticality of the examples in (23), as
Simpson points out. Chomsky (1995, 2000) asserts that the operation Merge, in which
an element is simply merged into the structure, is more economical than Move, the
composite operation which involves Agree, pied-piping, and re-Merge combined. A
sentence like (30) below is barred because in the derivation of the clause to be a man
in the garden, there are two choices if the numeration contains an expletive. The first
is to Merge the expletive there (later raising it further) as in (29). The second is to
Move the DP a man and postpone Merging the expletive, as in (30). (30) is
91
ungrammatical because the less economical operation was chosen at that point in the
derivation.
(29) There seems [t to be a man in the garden].
(30) *There seems [[a man] to be t in the garden]].
Returning to wh-expletives, if wh-expletives are simply merged into the specifier
of CP, this will always be more economical than moving a full wh-phrase. According
to Simpson (2000), the ungrammaticality of (31a-b) demonstrates that wh-expletives
are not simply merged into a wh-position.
(31) a. *Was glaubst du [was Hans wen gesehen hat]
Expl believe you expl Hans whom seen has
Intended: ‘Whom do you believe Hans saw?’ [German]
b. *Was glaubst du [ t Hans wen gesehen hat]
Expl believe you Hans whom seen has
Intended: ‘Whom do you believe Hans saw?’ [German]
If Merging of a wh-expletive was more economical than Moving the wh-phrase into
the lower Spec, CP, we would expect (31a-b) to be possible. However, if we accept
that was as a wh-expletive must be assigned case, it would not be possible to generate
it in the subordinate clauses in (31) in which all available cases have been assigned to
full wh-phrases. This is an extension of the account of (23).
Simply stipulating anti-locality as a requirement ("an expletive and any full wh-
phrase cannot be clausemates") cannot be adequate, since it is possible for a full wh-
phrase to appear in the same clause as the wh-expletive if the full wh-phrase is a
92
subject (nor would this requirement prevent ungrammatical sentences as in (31)). In
this case, the matrix wh-expletive indicates the position of interpretation of the wh-
phrase in the lower clause, and has nothing to do with the wh-phrase in the matrix
clause. This construction is exhibited in German by (32) (originally from Fanselow
and Mahajan (1996)) and in Hindi-Urdu by (33).
(32) Was glaubt wer wen sie liebt4
Expl believe who whom she loves
‘Who believes she loves whom?’ [German]
(33) kis-ne kya socaa ki aap-ne kya paRhe?
who expl thought that you what read
‘Who thinks you read what?’ [Hindi-Urdu]
As long as the wh-expletive is being assigned a DP case separate from that of
the full wh-phrase in the same clause the construction is grammatical.5
4 There has been some discussion about the grammaticality of (32). While Dayal (1994) claims that this sentence is ungrammatical, both Simpson (2000) and Beck and Berman (2000) indicate that this sentence is grammatical or nearly so for most speakers. 5 Sandy Chung has pointed out to me that we must also ask whether it is possible for a wh-expletive to co-occur with an interrogative subject or adjunct. This is ungrammatical in Kashmiri: (i) *k’a von kamI ki Sita chi sehatas manz. expl said who ki Sita is health in. Intended: ‘Who said that Sita is healthy?’ (ii) *tse k’a chu-y kati pata: ki Sita chi sehatas manz. You expl aux how know that Sita is health in. Intended: ‘How do you know that Sita is healthy?. This account correctly predicts that such sentences will be ungrammatical because of the properties of v. In the case of (i), v probes its domain and finds no interrogative material (besides the expletive with no interpretable features). Its [wh]-feature therefore cannot be valued. In (ii), assuming that both the wh-expletive and full wh-phrase originate in the domain of v, if the wh-expletive raises through the specifier of vP, the full wh-phrase kese will not be able to have its [Q-u] feature valued (it will be too far from C, and v can have no [Q] in this derivation or the wh-expletive would have been frozen in its specifier).
93
3.3 Previous Approaches to Wh-expletive Constructions: Indirect and Direct
Dependency
Indirect dependency approaches to wh-expletive constructions generally
follow the account of Hindi-Urdu first proposed by Dayal (1994, 1996). Although we
will address concerns particular to Hindi-Urdu in Chapter 4, it is essential to review
the indirect dependency account here as it is frequently cited as a way of accounting
for wh-expletive constructions in a range of languages (as recently as Rackowski and
Richards 2005, Bruening 2006). The core claim of the indirect dependency proposal
is that all apparently subordinate clauses in Hindi-Urdu are in fact adjunction
structures, and that what we have called a wh-expletive in this chapter is in fact a
scope-marker that is coindexed with the adjoined CP. The process of semantic
interpretation of the coindexed components proposed by Dayal allows the scope of
the wh-phrase in the adjoined CP to be interpreted at the position of the scope-
marker. Dayal referred to this as 'indirect dependency' to contrast it with chain-based,
'direct dependency' approaches which rely on syntactic connections between the
element kyaa (construed now as an expletive) and the full wh-phrase (these will be
discussed in greater detail below).
The indirect dependency approach is an attempt to address the following
puzzle: wh-in-situ phrases in a single Hindi-Urdu clause take scope over the entire
clause.
94
(34) tum kis-ko pasand karte ho
You who-acc like do aux
Whom do you like? [Hindi-Urdu]
According to Dayal, this indicates that the wh-phrase moves to Spec, CP at the level
of Logical Form (LF). In a two-clause question in which the wh-in-situ phrase is in
the lower clause, a matrix scope interpretation is not available.
(35) tum jaante ho [ki us-ne kyaa kiyaa]
you know aux that he-E what do
You know what he did.
≠ What do you know that he did? [Hindi-Urdu]
According to Dayal, this fact demonstrates that the embedded wh-phrase
cannot move to the Spec, CP of the matrix clause at LF. If this is the case, then there
must be some reason why this movement is blocked. Dayal claims that finite
subordinate clauses are adjoined to the matrix clause CP or IP. Support for this claim
is derived both from the scope facts in (35), and from constituent order in Hindi-
Urdu.
In a standard Hindi-Urdu clause, complements precede verbs in an unmarked
sentence.
(36) Hamid-ne pani piya
Hamid-erg water drank
‘Hamid drank water.’ [Hindi-Urdu]
95
However, finite complement clauses appear exclusively to the right of the verb.
(37) a. Vo jaantii hai ki anu aayii
She know aux that Anu come
‘She knows that Anu came.’ [Hindi-Urdu]
b. *Vo ki anu aayii jaantii hai [Hindi-Urdu]
If, as Dayal claims, all seemingly finite complement clauses in Hindi-Urdu are in fact
adjoined, these clauses could be considered strong islands. Following the definition of
barrier proposed in Cinque (1990), this would be because an adjoined clause is in a
position not directly selected by the verb and is in the non-canonical direction. It is
therefore a barrier for both binding and government. Further, if we follow Dayal’s
proposal that subjacency is operative at LF, it will be impossible for any in situ wh-
phrase to escape the finite complement clause. For this reason, it cannot take matrix
scope under any circumstances.
It then becomes a puzzle why the scope-marking structure in (38) permits a
matrix interpretation of the wh-phrase. The phrase should never be able to escape the
lower clause at LF.
(38) Sita-ne kyaa socaa ki Ravii-ne kis-ko dekhaa?
Sita-erg expl thought that Ravi-erg who-acc saw
‘Who did Sita think that Ravi saw?’ [Hindi-Urdu]
In these structures, Dayal claims that the complement position of the main verb is
occupied by a scope-marker (kyaa), what we have called a wh-expletive to this point.
The scope-marker and the adjoined CP are coindexed. Since Hindi-Urdu is an SOV
96
language, in this approach the pre-verbal position that the wh-expletive occupies
appears also to be the canonical object position. According to Dayal, in Hindi-Urdu
all in situ wh-material must raise to the edge of its CP at LF, creating two local wh-
dependencies in a two-clause partial movement construction. The first is due to the
LF movement of the full wh-phrase to the specifier of the adjoined CP, and the
second to the LF movement of the scope-marker kyaa to the specifier of the matrix
CP. Since the scope-marker and the adjoined CP are related by coindexation, the net
result gives the effect of a single long-distance dependency. The LF structure this
process would produce for (39a) is represented in (39b).
(39) a. sita-ne kyaa socaa ki ravi-ne kis-ko dekhaa
sita-erg expl thought that Ravi-erg who-acc saw
‘Who does Sita think Ravi saw?’
b. [CP
kyaa1 [ [
IP sita-ne t
1 socaa] [
CP ki kis-ko2 ravi-ne t
2 dekhaa]
1 ]
This view achieves the effect of interpreting the full wh-phrase on the left periphery
of the matrix clause because that is where the entire adjoined CP will be interpreted.
Dayal extends this approach to German question formation, proposing a nearly
identical LF for German wh-expletive constructions.
The indirect dependency approach raises a number of questions, both
theoretical and empirical. Perhaps the most unusual claim of this approach is that
what is normally viewed as a complement CP is in fact adjoined to the matrix CP. If
97
this is the right view, we should expect that interactions which depend on command
relations into the rightmost clause will work in very different ways than in languages
in which that clause is a complement to V (as is presumably the case in English). For
instance, it should not be possible for quantifiers in the matrix clause to bind variables
in the adjoined CP, since under this view these quantifiers would not c-command the
variables. However, in Hindi-Urdu wh-expletive constructions like (40) it appears
that the quantifier har aadmii 'each man' in the matrix clause in fact can bind the
pronoun us-ne 'he' in the second CP (Mahajan, 2000).
(40) Har aadmii-nei kyaa socaa ki us-ne
i kis-ko dekhaa.
each man-erg expl thinks that he-erg who saw
‘Who did every mani think that hei saw?’ [Hindi-Urdu]
Under the indirect dependency approach, the bound variable reading in (40) is
unexpected. Similarly, in (41) a matrix clause complement binds a pronoun in the CP
that follows the verb bola ‘told’. This bound variable reading would be surprising if
the CP is adjoined higher than the object, since the object could not c-command the
pronoun inside CP.
(41) aap-ne har aadmi-sei kyaa bola ki voi fon-par kis-se bula: sakta he
You-erg each man expl told that he phone-on who call can aux
Who did you tell each man that he could telephone? [Hindi-Urdu]
98
Along these same lines, Beck and Berman (2000) point out that if German
clauses containing expletives was and es cause their complements to be adjoined,
then the binding in (42) should be impossible.
(42) daß keine Studentini es bedauert, daß sie
i die Vorlesung geschwänzt hat
that no student it regrets that she the lecture skipped has
‘... that no student regrets it that she has skipped the lecture.’ [German]
Again, since the embedded clause would hypothetically be adjoined to the first CP,
the matrix subject should not c-command any material within it, and this binding
would be ruled out.
On a more theoretical note, the origin of the coindexation of the wh-
expletive (and the free variable it is translated as) with the embedded wh-clause is
unclear. There appears to be no motivation beyond achieving the correct
interpretation of the adjoined CP since, as Beck and Berman point out, the indexation
is neither referential nor anaphoric. Further, if we are to explore the viability of the
Strong Minimalist thesis for language, the Inclusiveness Condition requires that no
new features be introduced in the course of a derivation (Chomsky 2000, p. 113).
Indices and coindexation become unavailable in this view. Of course, this condition
would also be problematic for incarnations of the direct dependency approach that
rely on syntactic chains established between wh-items to arrive at the appropriate
connections within the structure. We will return to the indirect dependency approach
99
below addressing its similarities and differences with the account that will be offered
here.
The account proposed in this chapter follows the indirect dependency
approach in assuming that the wh-expletive is base generated in a position inside the
clause in which it is assigned case. This claim was supported by evidence of various
kinds in section 3.2. However, it departs from the indirect dependency view in that
the role of the wh-expletive is taken to be syntactic only, and the wh-expletive does
not play a role in the semantic interpretation of the question (for more on this point,
see section 3.4).
It may seem, then, that the account presented here has more in common with
direct dependency approaches, but this is probably misleading. The representative
direct dependency approach to partial wh-movement and wh-expletive constructions
is McDaniel's (1989) account of German and the Indic language Romani. In this
account, a syntactic wh-chain with a specific set of well-formedness conditions
connects a wh-phrase and its trace with a wh-expletive. McDaniel claims that the
unique properties of wh-expletive constructions can be explained if Subjacency is
considered to be a condition on representations (not just movement). This type of
analysis seems especially geared toward wh-expletive constructions of three clauses
or more, as discussed in 4.2 below.
The direct dependency approach attempts to codify the notion that full and
partial wh-movement are different manifestations of the same phenomenon. Under an
account of this kind, both full movement and wh-expletive constructions result in
100
chains with similar (though not identical) well-formedness conditions. This is a view
shared by the approach proposed here, though it is encoded in a very different way. In
this case each head in a series of clauses in a matrix question must have its features
valued and EPP satisfied. This can be accomplished either via movement, or via wh-
expletive insertion, depending on the numeration; the end result is essentially the
same. In the indirect dependency approach, on the other hand, full wh-movement
constructions are a syntactic phenomenon, while wh-expletive constructions are not.
That is, the crucial role played by a wh-expletive itself is a semantic one. These two
question formation strategies are viewed as totally distinct. The claim, on the other
hand, that both full wh-movement and wh-expletive constructions are manifestations
of the workings of the same set of mechanisms is something that the direct
dependency account and the account presented here share.
In the approach in 3.1, a connection is made in the syntax between the
position at which the wh-phrase will be interpreted and the wh-phrase itself. It is the
features of the full wh-phrase in a sentence like (2) which value the uninterpretable
features of the head in the matrix position where the wh-phrase will be interpreted.
Unlike the direct dependency view, however, there is no link made between the wh-
expletive and the full wh-phrase. The connection that is forged makes use of neither
chain nor index and is effected in the course of normal syntactic processes, requiring
no additional mechanisms beyond featural satisfaction and associated movement.
This feature-based approach permits an important simplification over chain-
based direct dependency approaches. McDaniel's account requires a stipulation
101
contained in the well-formedness condition on chains to ensure that wh-expletives
appear only above the full wh-phrases whose scope they indicate in syntactic
structure. This is stated in (43):
(43) for any scope-marker ai , 1 <i <n, (a
i+1 , . . . , a
n-1 ) contains a true wh--
phrase (McDaniel 1989, p. 580)
In the account presented in section 4, it follows automatically that wh-expletives will
only occur in clauses higher than the one hosting the wh-phrase whose scope they
mark. More precisely, the feature-based system requires that something with a wh-
feature appear in the domain of the lowest interrogative C-head. This is because this
head possesses an uninterpretable [wh] feature. Wh-expletives do not have a [wh]
feature at all (by definition), so if a wh-expletive is the only wh-material to appear in
the domain of the lowest interrogative head, the uninterpretable [wh] feature on that
head will remain unvalued. In this case the derivation will crash. This result is
achieved without additional stipulation.
3.4 Interpreting wh-expletive constructions
This section is intended to suggest an approach to the semantic interpretation
of wh-expletive constructions that is compatible with the syntactic analysis presented
above. The requirement imposed on such an approach is that the wh-expletive, which
has no interpretable features at all, plays no role in the matrix interpretation of
embedded wh-phrases. Only those elements of the A-bar system with interpretable
102
features can be relevant for interpretation, since all uninterpretable features are
deleted by the conclusion of the derivation. These elements are the matrix C-head,
which hosts an interpretable [Q] feature, and the wh-phrase itself, which possesses an
interpretable [wh] feature.
Reinhart (1998) proposes a mechanism for binding wh-in-situ constituents
that is meant to be compatible with the general architecture of the Minimalist
Program. In that vein, it seeks to eliminate the need for movement at LF to achieve
wide scope, and instead proposes that these constituents must be interpretable in-situ.
I will not attempt a full development of this analysis; I only want to sketch how a
reasonable semantics can be built on the syntactic foundation developed here.
Reinhart follows Karttunen (1977) in assuming that a wh-NP is simply an
indefinite NP, and a question denotes the set of (true) propositions that are its
answers. She claims that the question operator (at the position at which the scope of
the question will be interpreted) introduces an existential quantification over a choice
function variable. The wh-word, and in fact any weak determiner, acts as a (choice)
function variable applying to the set denoted by the NP. This variable can be bound
by an existential operator that may be arbitrarily far away. In this way, wide-scope
interpretation can be assigned to in-situ wh-constituents. Note that on this view, no
LF-raising of the wh-phrase is required. In fact, talk of the “scope” of the wh-phrase
is somewhat misleading.
This proposal dovetails well with the syntactic approach put forward above. In
this view, the interpretable [Q] feature on the matrix (or highest relevant) C-head
103
triggers as existential quantification over a choice function variable. The interpretable
[wh] feature reflects the fact that the wh-indefinite is interpreted as a choice function
variable that applies to a given set. For the wh-expletive construction in (2), repeated
here as (44), the relevant portion of the interpretation is sketched in (45).
(44) tse k'a: chu-y ba:sa:n ki mohn-as kem' dits kita:b
you expl aux think that Mohan who gave book
‘Who do you think gave Mohan the book?’
(45) {P (∃f) (CH(f) & you think that f(xanimate(x)) gave Mohan the
book}6
In this case, the [Q-i] in the matrix clause provides existential quantification over a
choice function, binding the function variable provided by the [wh-i] of the wh-NP7.
This question then denotes the set of true propositions P, each stating that for some
choice function f that you think that the animate being selected by f gave Mohan the
book. This binding can occur over an arbitrarily long distance, so it should not matter
how deeply embedded the full wh-phrase is.
Our syntactic proposals are thus compatible with a reasonable semantics –
one that was developed on entirely independent grounds. Another kind of question
now arises: if the wh-expletives are irrelevant for effecting the matrix scope
6 Reinhart assumes that pronominal wh-phrases such as who are determiners with empty nouns: [who
[Ne(i)]] 7 The interpretable [Q] feature also seems to be what establishes the set of propositions.
104
interpretation of the wh-phrase, why are they present? The analysis presented above
makes clear that the role of the wh-expletive is purely syntactic in operational terms.
It fulfills the syntactic requirement we have labeled with the name EPP – that certain
heads require additional specifiers beyond those needed for selectional purposes. The
appearance of these wh-expletives is not a semantic requirement, as the system of
interpretation has no need for intermediate material to effect the matrix scope
reading8. The larger question – of why the design of natural language includes
expletive elements – of course remains.
4. Remaining Issues
There are two remaining issues concerning wh-expletive constructions that
remain to be discussed. The first question is a Kashmiri-internal one, and its solution
does not bear significantly on the analysis presented above. The second question is
crosslinguistic in nature, and has ramifications for the wider study of wh-expletives
across languages and dialects.
4.1 A Kashmiri Issue: Factive Predicates
Bhatt (1999) claims, and my informants agree (PK 9/21/04, JC, VC 9/8/05),
that full wh-movement is not possible from clauses that follow non-bridge verbs in
8 This will be supported by certain varieties of wh-expletive languages which have three clause structures lacking wh-expletives in the intermediate clause, discussed in section 4.2.
105
Kashmiri, but wh-expletive constructions are generally permitted (compared with the
grammatical bridge verb versions of these sentences in (1)-(2)).
(46) a. *k'a: cha mi:ra:yi khabar ki t mohnan por.
what aux Mira know that Mohan read
Intended: ‘What does Mira know that Mohan read?’
b. k'a: cha mi:ra:yi khabar ki k'a: mohnan por.
expl aux Mira know that what Mohan read
‘What does Mira know that Mohan read?‘
Others (namely Wali and Koul 1997) have claimed that wh-expletive constructions
are also impossible in factive contexts, and that (46b) is ungrammatical. It is clear that
there is further empirical work to be done here, but presuming for the moment that the
judgments in (46) are correct for at least some varieties of Kashmri, we can explore
several ways of accounting for them.
Bhatt (1999) has proposed that subordinate clauses out of which extraction is
impossible in Kashmiri are in fact adjoined to the matrix clause. The entire structure
would be as in (47).
(47)
CP CP CP mi:ra:yi cha khabar ki k’a: por mohnan Mira aux knows that what read Mohan.
106
In this view, wh-movement will not be possible between the two clauses because the
first CP does not contain the second. There is, however, an approach to the structure
in (47) in which k'a: could be viewed as marking the scope of the wh-phrase in the
adjoined clause. This approach, along the lines of Dayal (1996), was already
addressed in section 3.2. Suffice it to say here that under such an "indirect
dependency" approach one could obtain the contrast in (46).
On the other hand, binding facts in Kashmiri once again indicate that a
hierarchical relationship obtains between the matrix and complement clauses of both
factive and non-factive predicates (JC 6/05).
(48) a. har insa:nasi chu basa:n ki su
i chu te:z.
each man aux thinks he aux smart
‘Each man thinks he is smart.’
b. har insa:nasi chu pata: ki su
i chu te:z.
each man aux knows he aux smart
‘Each man knows he is smart.’
In both sentences in (48), har insa:nas binds the pronoun in the lower clause,
indicating that it commands the pronoun. This is unexpected if the structure of the
factive sentence was as in (47).
An alternative solution to this puzzle could be related to the fact that factive
verb complements are weak islands in other languages.
(49) a. How do you think he escaped the building?
107
b. *How do you regret he escaped the building? (on matrix reading)
Assuming the wh-expletive construction to be an instance of the operation
Agree (simplex), Agree may be able to obtain across weak island boundaries even
though Movement (complex) is impossible across those same boundaries. Clearly this
notion needs further work and formalization, but in light of the facts in (48) it seems
more promising than an adjunction account. Moreover, the controversial nature of the
data as it is reported in the literature indicates that further empirical investigation is
required.
4.2 A Crosslinguistic Issue: Multiple wh-expletives
The crosslinguistic issue that remains to be examined is centered on wh--
questions consisting of more than two clauses. In German, Hindi-Urdu, and Kashmiri,
an embedded wh-phrase can take scope across any number of clauses so long as this
is mediated by a wh-expletive in every clause higher than the clause containing the
wh-phrase.
(50) a. Was glaubst du, was Jan meint, mit wem Ann gesprochen hat
Expl believe you expl Jan think, with whom Ann talked has
‘Who do you believe Jan thinks Ann talked with?’ [German]
b. *Was glaubst du, Jan meint, mit wem Ann gesprochen hat?
(51) a. Ram-ne kyaa socaa ki ravii-ne kyaa kahaa ki kon sa aadmii aayaa
Ram expl think that Ravi expl said that which man came
‘Which man did Ram think that Ravi said came?’ [Hindi-Urdu]
108
b. *Ram-ne kyaa scoaa ki ravii-ne kahaa ki kon sa aadmii aayaa
(52) a. Raman k'a: von ki tse k'a: chu-y ba:sa:n ki mohn-as kem' dits
Ram expl said that you expl aux think that Mohan who gave
kita:b?
book
‘Who did Ram say you think Mohan gave the book to?’ [Kashmiri]
b. * Raman k'a von ki tse chu-y ba:sa:n ki mohn-as kem' dits kita:b
In (50a), (51a), and (52a) the wh-phrase has undergone partial movement to the left
periphery of the lowest clause. A wh-expletive is present in both the intermediate and
matrix clauses, and a matrix interpretation of the wh-phrase results. The
ungrammatical examples, (50b), (51b), and (52b) are identical, except that they lack a
wh-expletive in the intermediate clause. It seems that a wh-expletive must be present
in every clause above the clause hosting the wh-phrase, up to and including the clause
in which the wh-phrase is bound by the interrogative operator.
The first question is how to account for the core pattern: the obligatory
presence of an expletive in each clause above the clause containing the wh-phrase.
The partial movement of the wh-phrase in examples like (52b) is explained just as it
is in two-clause structures: in this way the EPP property of the lowest C-head is
satisfied, and the wh-phrase becomes accessible to higher probes which will value its
109
remaining features. If the wh-phrase remains in this low position however, the EPP
on all higher probes must be satisfied by a wh-expletive. If an expletive is missing in
any of these positions, the EPP will not be satisfied on the C-head in that clause.
Upon insertion of an expletive, this C-head will still have features that need to be
valued by Agree. It can do so by probing the lower phase edge and interacting with
features found there. In what follows I will make this account more explicit.
Consider the three possible ways of forming a three-clause wh-question in
Kashmiri in which the wh-phrase originating in the lowest clause takes scope over the
entire sentence (Wali and Koul 1997, Bhatt 1999, JC & VC 9/8/05) .
(53) Raman k´m von ki tse chu-y ba:sa:n ki mohn-as dits kita:b?
Ram who said that you aux think that Mohan gave book
‘Who did Ram say you think Mohan gave the book to?’ [Kashmiri]
(54) Raman k'a von ki tse k´m’ chu-y ba:sa:n ki mohn-as dits kita:b?
Ram expl said that you who aux think that Mohan gave book
‘Who did Ram say you think Mohan gave the book to?’ [Kashmiri]
(55) Raman k'a von ki tse k'a: chu-y ba:sa:n ki mohn-as k´m' dits kita:b?
Ram expl said that you expl aux think that Mohan who gave book
‘Who did Ram say you think Mohan gave the book to?’ [Kashmiri]
(53)-(55) illustrate the three different ways to ask the matrix question ‘Who did Ram
say you think Mohan gave the book to?’. In (53) we have full wh-movement from the
base position in the lower clause (indirect object of dits ‘gave’), to the canonical wh-
110
position in the matrix clause, preceding the second-position verb. In (54), the wh-
phrase has moved up only one clause, to the preverbal position in the intermediate
clause, and a wh-expletive occupies the wh-position in the matrix clause. Below I will
demonstrate that this case, not surprisingly, reduces to the analysis given to two-
clause wh-expletive constructions. In (55), the wh-phrase has moved only to the left
periphery of its own clause. A wh-expletive occupies the wh-position in both the
intermediate and matrix clauses.
In the case of full wh-movement, as found in (53), this approach would offer
the relatively straightforward account outlined in the diagram in (56).
(56) [wh-phrase [C1 …][t [C2] …][t [C3] … t]
[Q-u] [Q –i] [wh-u] [wh-u]
[wh-i] [wh-u] [EPP] [EPP]
[EPP]
The wh-phrase that originates inside the lowest clause is an active goal. The C-head
on the left periphery of this clause (C3) is an active Probe. The wh-phrase raises into
its specifier, valuing its uninterpretable [wh] feature and satisfying the EPP. As it is
on the edge of the lowest phase, it is visible for interactions in the next phase up.
Furthermore, it still has unvalued features (since C3 is non-interrogative and cannot
value its [Q] feature). Therefore, it can function as an active Goal. The process is then
repeated in the phase determined by C2. The wh-phrase at this point is still an active
goal because its uninterpretable [Q] feature has not been valued by C2 either. The
process is repeated in the phase determined by C1, but on this step, when the wh-
111
phrase raises into the specifier of C1, its [Q] feature is valued. At this point, all
uninterpretable features have been valued, and the derivation is licit.
The sentence in (54) is a mixed case, in which a part functions just like the
full wh-movement described above, and a part functions like a wh-expletive
construction. This process is diagrammed in (57).
(57) [wh-expl [C1 …t ][wh-phrase [C2] …][t [C3] … t]
[Q-u] [Q –i] [Q-u] [wh-u] [wh-u]
[wh-u] [wh-i] [EPP] [EPP]
[EPP]
In (54), the wh-phrase raises from the lowest to the intermediate clause in the same
manner described above in (56). However in (57), a wh-expletive is present in the
numeration and is introduced in a case position inside the matrix clause. C1 is an
active Probe and finds the wh-expletive that is an active Goal. The wh-expletive
moves into the specifier of C1, and its single remaining uninterpretable feature is
valued. The EPP on C1 is satisfied, but is own uninterpretable [wh] feature has yet to
be valued. C1 probes its domain once more and finds the full wh-phrase at the left
edge of the lower phase. The wh-phrase is still an active goal, because its [Q-u] is still
unvalued. C1 and the wh-phrase Agree with one another, mutually valuing one
another’s uninterpretable features and creating a licit derivation. The second part of
this derivation is identical to the way in which a two-clause wh-question is analyzed.
A third possible way to form a three-clause matrix question in Kashmiri is to
permit the wh-phrase to remain in the lowest clause by employing wh-expletives in
112
both the intermediate and matrix clauses, as in (55). The analysis that has been
described here would give (55) the structure described by the schematic in (58).
(58) [wh-expl [C1 …t ][wh-expl [C2] … t ][wh-phrase[C3] … t]
[Q-u] [Q –i] [Q-u] [Q-u] [Q-u] [wh-u]
[wh-u] [wh-u] [wh-i] [EPP]
[EPP] [EPP]
Let us walk through each step of (58) carefully. The initial step of (58) is identical to
that in (56) and (57), in which the wh-phrase moves from its base position into the
specifier of the lowest C-head. The EPP on C3 is satisfied, and its single
uninterpretable feature is valued. The wh-phrase, on the other hand, remains active. In
the intermediate clause, the C2 head probes its domain and finds the wh-expletive.
The expletive moves into the specifier of C2 and satisfies the EPP while valuing its
own uninterpretable [Q] feature. Though this wh-expletive is now inactive, the C2
head still has an unvalued uninterpretable [wh] feature. It probes its domain once
again and interacts with the wh-phrase at the left edge of the phase below. The C2
head and the wh-phrase mutually value one another’s features. The derivation
continues in the matrix clause, in which the C1 head interacts with the wh-expletive
originating in its own clause, and the process proceeds as above. The C1 head must
then probe its domain again, as it is still active, and find the valued [wh] feature on
the C2 head with which to value its own [wh] feature. At this point all uninterpretable
features have been valued, and the derivation is licit.
113
Notice that the wh-expletives in (58) do not interact with the wh-phrase in any
way. In fact it is the C-heads in each clause that interact with the wh-phrase, not the
wh-expletive. The role of the expletive is solely to satisfy the EPP – it has no other
purpose. This approach permits an arbitrarily long series of C-heads to value one
another via static Agree over a distance. In this view, there need be no direct link
between the expletives and the full wh-phrase. The wh-expletive does not possess
interpretable features, so will ultimately be meaningless, and only the features of the
full wh-phrase will be relevant in the interpretation of the question.
The analysis of long-distance wh-expletive dependencies presented in (58)
captures an important intuition about the construction in (55). There is a sense in
which both wh-movement and wh-expletive constructions are alternative strategies
for forming a long-distance wh-dependency. In this approach to wh-expletive
constructions, it is the features of the wh-phrase itself that cause features on the
successive C-heads to be valued. In (58), it is the features of the wh-phrase that value
the intermediate C-head, and those features which in turn value the features of the
matrix C-head. Some information about the wh-phrase is affecting the series of heads,
passed up from one clause to the next. In this way, this account shares with direct
dependency approaches the notion that wh-movement and wh-expletive constructions
are two methods of creating the same long-distance wh-dependency. Here, it is the
features of the full wh-phrase that ultimately cause the features of the matrix C-head
to be valued, whether it moves there itself, or whether a wh-expletive fills the position
so that static Agree can occur.
114
Let us turn finally to a potentially troubling aspect of these proposals. In (58),
the intermediate C-head, like the matrix C-head, has a [Q] feature. This causes the
wh-expletive in its specifier to have all its uninterpretable features valued, and thus be
frozen in place. We can now note that this has an additional consequence, of valuing
the uninterpretable [Q] feature on the wh-phrase in the specifier of the lowest C-head.
In effect, this makes the lower two clauses of the structure in (58) analogous to any
two-clause wh-expletive construction. The analysis in (58) ensures not only that the
wh-expletive is frozen in place in the specifier of the intermediate C-head, but also
that the wh-phrase in the lowest clause is frozen in the lowest C specifier. There is
crosslinguistic evidence that intermediate heads in a wh-movement sequence may
possess interrogative features of some kind. In particular, Henry (1995) observes that
in Belfast English subject-auxiliary inversion takes place not only in the highest C
head in a wh-movement sequence, but in intermediate heads as well. She takes this as
an indication that at least in this context these intermediate heads have some
interrogative status just like matrix C-heads (see also Rizzi 1996). In the present
account, this notion is reflected in the [Q] feature which appears on some
intermediate C-heads in Kashmiri.
A central element in these proposals (evident in (58)) is the availability of
three different flavors of C-heads – each with a slightly different set of interpretable
and uninterpretable features. A legitimate concern arises at this point: what if it is not
these exact combinations of features that appear on these heads in the numeration and
in this just this order? For instance, what if an intermediate C-head in a three clause
115
wh-expletive construction was not of the type containing a [Q-u]? (Of course, the
presence of a [Q-i] at the top of an interrogative clause is going to be mandated by the
selectional requirements of the governing predicate). One way of answering this
question is to work through this and other scenarios for both wh-movement and wh-
expletive strategies (as is done in the Appendix). However the larger answer to this
family of questions is that derivations and numerations with flaws such as these fail to
converge. Some uninterpretable feature in each case will remain unvalued and will
result in a crash. The system of features proposed here is such that heads which must
trigger a certain semantic interpretation (for matrix questions, the matrix C-head)
possess certain interpretable features. Uninterpretable features on those heads serve to
enact movement of and agreement with various goals. Certain sets of features
appearing on heads in a certain order limit the patterns that can appear. Other patterns
will result in non-convergent derivations. As opposed to a list of maxims, principles,
and conditions, in this framework the properties of the grammar are expressed via
features and the systematic operations that result in their valuation. Though we will
not explore this issue here, it should be noted that it is part of an important set of
broader questions about the way in which the Minimalist Program accounts for
ungrammaticality.
Through (56)-(58), the feature-based approach to wh-expletive constructions
in 3.1 has been extended to account for such constructions in sentences of three
clauses (in fact to sentences of arbitrarily many clauses). Essential to this account are
two operations that can function alone or in concert. The first is interclausal wh-
116
movement, in which a wh-phrase moves from one clause to the next through a series
of C-heads. These C-heads cannot value the uninterpretable feature on the wh-phrase,
so the wh-phrase must continue until reaching the matrix clause. The second
operation is wh-expletive clause internal movement, in which the C-head has the
capacity to deactivate the wh-expletive by valuing its single uninterpretable feature.
The wh-expletive only values the EPP on this type of C-head, and then remains fixed
in its specifier. The C-head then acts as a probe and values its uninterpretable features
via static Agree within its domain. An example like (53), with long wh-movement,
makes use only of the first operation described here. (54) and (to a lesser extent) (55),
make use of both of these operations in combination. Given the different flavors of
functional heads that can appear, and the different types of wh-elements in languages
like Kashmiri (meaningless wh-expletives and meaningful wh-phrases), it is
predictable that there will be several ways to form a matrix question in a three-clause
sentence.
At this point, however, an unresolved issue arises concerning (52b). An
alternative derivation of (52b) is conceivable in which the EPP of the intermediate
clause is satisfied. If the wh-expletive that appears in the matrix clause originated in
the intermediate clause and subsequently moved higher, then we would need to find
an alternative explanation for the ungrammaticality of (52b). Such a derivation would
be one in which the wh-expletive in the intermediate clause raised into the specifier of
the intermediate C-head, then raised further into the specifier of the matrix C-head. In
117
this case the EPP would be satisfied in both positions, and (52b) should not be
ungrammatical.
This would be analogous to the behavior of DP-expletives. Consider the English
DP-expletive there in (59).
(59) There seems to be a problem.
The standard account of (59) is that there originates in specifier of the TP [to
be a problem], and then raises into the specifier of the higher TP. However, there are
several reasons to believe that wh-expletives, unlike DP-expletives, cannot move
from one clause to another. I will address these reasons below, and will ultimately
conclude that the ungrammaticality of the (b) sentences above can be explained as a
failure to satisfy the intermediate EPP.
Part of the evidence for rejecting wh-expletive movement comes from
Hungarian. Recall above that the explanation for the so-called crosslinguistic anti-
locality effect of wh-expletives was that wh-expletives originate in a case position
inside their clause. Therefore sentences in which the wh-expletive appears in the
same case-domain as a full wh-phrase will be correctly predicted to be impossible
because of “case-competition” between the expletive and the wh-phrase. In
Hungarian, the base-generated position of the wh-expletive is quite clear, since it
bears a morphological indication of the case is assigned.
Consider the three-clause question in (60). The verb in the intermediate
clause, szamitasz, assigns allative case to the wh-expletive in its clause, resulting in
118
the form mire. The matrix verb hitt assigns accusative case to its wh-expletive,
resulting in the form mit.
(60) Mit hitt Janos, hogy mire szamitasz, hogy mit fognak
expl-acc thought-indef Janos that expl-al count-2sg that what will
mondani a gyerekek? [Hungarian]
say-inf the kids-nom
‘What did John think that you expect that the kids will say?’
(Simpson 2000)
Adding complexity to this question, Horvath (2000) and Simpson (2000) report that
sentences of the type in (50b)-(52b), in which the intermediate clause lacks a wh-
expletive, are possible in Hungarian, as well as in some varieties of German and
possibly dialects of Kashmiri (these varieties will be discussed in detail below). If
these sentences were formed via movement of the intermediate wh-expletive into the
highest clause, we would expect that the sole wh-expletive in the sentence would be
the allative form mire. This prediction is not borne out.
(61) Mit/ *mire hitt Janos, hogy szamitasz, hogy mit fognak
expl-acc/expl-al thought-indef Janos that count-2sg that what will
mondani a gyerekek? [Hungarian]
say-inf the kids-nom
119
‘What did John think that you expect that the kids will say?’
(Simpson, 2000)
Instead, it is the accusative form of the expletive mit that is grammatical, as is
apparent in (61). If the expletive that ultimately appears only in the matrix clause
originated in the intermediate clause, it would first have had to receive allative case
from szamitasz, then accusative case from hitt. It seems clear that this wh-expletive
was base-generated in a case position in the matrix clause. Simpson takes this to
indicate that wh-expletives cannot move from one clause to another.
We can also find support for this position in a variety of Kashmiri (let us call
it Kashmiri B) which permits sentences like (52b). Let us compare the sentence with
a wh-expletive in each clause, with the sentence that lacks a wh-expletive in the
intermediate clause (repeated here as (63), and marked grammatical).
(62) Raman k'a: von ki tse k'a: chu-y ba:sa:n ki mohn-as k´m' dits kita:b?
Ram expl said that you expl aux think that Mohan who gave book
‘Who did Ram say you think Mohan gave the book to?’ [Kashmiri]
(63) Raman k'a: von ki tse chu-y ba:sa:n ki mohn-as k´m' dits kita:b?
Ram expl said that you aux think that Mohan who gave book
‘Who did Ram say you think Mohan gave the book to?’ [Kashmiri B]
[JC, VC (9/8/05), but PK (9/21/04) disagrees with this judgment]
In all dialects of Kashmiri, the pronoun preceding the wh-expletive in (62), tse ‘you’,
is interpreted as a topic. Any argument to the left of a wh-phrase in a Kashmiri clause
120
is a topic and would not, for instance, be able to host the focus-marker –ti (Bhatt
1999). In the sentence in (63), on the other hand, tse is interpreted as focused
material. It can grammatically host the focus-marker, and cannot be interpreted as a
topic [JC, VC 9/8/05]. This distinction would be unexpected if the wh-expletive in the
matrix clause in (62) had originated in the intermediate clause and then raised. If this
were the derivation of (63), the wh-expletive would have valued the interrogative
focus features on C (and no other focus features may simultaneously be present
according to the Kashmiri lexicon). This means that no other focus elements could
appear (Bhatt 1999). Tse would have to be located in the topic position above the first
specifier position of C. Movement of the wh-expletive to the matrix clause should not
change the informational role of tse. For this reason (63) serves as further evidence
that wh-expletives do not move out of the clause in which they originated.
Under the feature-based analysis presented for wh-expletives in two-clause
questions, it is expected that wh-expletives cannot move out of their clause. This is
because the wh-expletive has a single uninterpretable [Q] feature. The feature is
valued when the wh-expletive moves into the specifier of the C-head in the clause in
which it originates because in intermediate clauses in which expletives are generated
there is an uninterpretable [Q] feature on the C-head. Since the intermediate C-head
in these constructions values all uninterpretable features on the wh-expletive, then the
immobility of wh-expletives is accounted for. The wh-expletive becomes inactive and
is ‘frozen’ in place, unable to participate in any further interactions with higher
Probes. In this way, the system of features developed above to account for wh-
121
expletive constructions predicts that wh-expletives cannot move from clause to
clause.
Although this question is by no means resolved, I will for now conclude that
wh-expletives may not undergo further raising after they raise to the specifier of a
[Q]-bearing head and are thereby rendered inactive. Given this, we can assume that in
the ungrammatical (b) sentences in (50)-(52) the intermediate wh-expletive is simply
missing, and that each will fail due to the unsatisfied EPP in the first embedded
clause.
As is mentioned above, in Hungarian and some varieties of German and
Kashmiri, structures like the (b) sentences above are grammatical. In these sentences
there is apparently no expletive in the intermediate position. Simpson attributes this
pattern to phonological deletion of the intermediate expletive, and not expletive
movement. However, (63) suggests that this explanation is not plausible. There is
only one set of focus features possible on a C head in the lexicon of Kashmiri, and in
the intermediate clause in (63) the pronoun tse ‘you’ has interacted with these
features. This means that it is impossible for there to be another element, for instance
a wh-expletive, in this position at all, regardless of whether or not it is pronounced.
Although this question is not yet resolved, importantly the core pattern, in which the
wh-expletive obligatorily appears in all intermediate positions in a multiclausal
matrix question, does follow from the account presented here.
Throughout this discussion, we have built an account of A-bar movement that
is guided by the same general principles which underlie A-movement. The question
122
then becomes why expletive movement is possible in the A-system, but not in the A-
bar system. Consider once again (59), repeated here as (64).
(64) There seems to be a problem.
What is different about the DP-expletive there that permits it to move, while wh-
expletives may not? My tentative answer to this question here is: nothing. In fact, the
DP-expletive and the wh-expletive are roughly equivalent in terms of feature content,
each possessing only uninterpretable features. The difference lies instead in the types
of heads which exist in the A and A-bar systems. In the A-system, the non-finite T
head is often referred to as ‘defective’. It has no ability to assign case to a DP or even
to fully Agree with its phi-features. In the A-bar system, there is no analogous
‘defective’ head. Although there are different types of C-heads, all of them have
uninterpretable [wh] features which are valued by some [wh] feature in their domain.
So under this account there is no A-bar equivalent of the ‘defective’ T head. We can
simultaneously make an empirical observation that there is also no visible equivalent
of non-finiteness in the A-bar system – that is, there is no morphological realization
of ‘defectiveness’ like the English word to. For this reason I will assume here that the
fact that wh-expletives cannot move from one clause to another does not represent a
major split between the A and A-bar systems, nor does it represent a strong
distinction between the two types of expletives. It is also relevant to note here that
some researchers have concluded that A-expletives also do not undergo movement
(Bošković 2002), and that the reason for believing in A-expletive raising may only be
as strong as believing that the raising T head bears the EPP. Although more research
123
is certainly needed on this point, I will maintain that this supposed distinction does
not represent a major difference between A and A-bar expletives themselves.
5. Conclusion
Current theoretical frameworks provide us with a new way to approach wh-
movement and wh-expletive constructions. The account presented here claims that A-
bar movement, like A-movement, is driven by a system of interpretable and
uninterpretable features. This permits a new view of wh-expletive constructions in
languages like Kashmiri: wh-expletives serve to satisfy needs of certain heads in the
A-bar system, allowing those heads to interact at some distance with contentful wh-
material.
While this account covers significant empirical ground, it makes no use of
devices particular to the construction under analysis here. Instead the mechanisms
proposed apply to all of A-bar movement, and in fact reveal the symmetry of design
between the A and A-bar systems.
124
Appendix:
From the diagrams in (56)-(58) we see that there are three possible
interrogative feature bundles that can appear on a C-head in a three-clause matrix
question in Kashmiri. The presence of the feature bundle that contains [Q-i] is
determined by the selecting predicate (or in matrix position). However, other
predicates (such as the equivalents of think or say) could in principle select a C-head
with either of the two remaining bundles. These two bundles differ in that one
contains [Q-u] and one does not. The following schematics depict the three C-heads
in a three-clause matrix wh-question in which the full wh-phrase originates in the
lowest clause. They illustrate that if the incorrect feature bundle is chosen on one of
these two lower heads, the derivation will nearly always fail to converge.
The two patterns already established above in (56)-(58) are shown in in (i)-
(ii)9.
(i) C1 C2 C3 (see (56)-(57))
9 Clearly this list of features is not exhaustive; only the relevant features are displayed for the sake of
illustration.
125
wh-u wh-u wh-u
Q-i EPP EPP
EPP
(ii) C1 C2 C3 (see (58))
wh-u wh-u wh-u
Q-i Q-u EPP
EPP EPP
The feature bundles on the C-heads in (i) permit full wh-movement to the specifier of
C1, or partial wh-movement to the specifier of C2 with a wh-expletive in the specifier
of C1, but do not permit partial wh-movement only to the specifier of C3, because the
[Q] feature on the full wh-phrase could not be valued (there is no other [Q] feature on
an accessible Probe). The features in (ii) permit only partial wh-movement, to either
the specifier of C2 or C3, with wh-expletives in the specifiers of higher C-heads.
The patterns in (iii) and (iv) are not discussed in 4.2. In (iii) the C3 head
possesses Q-u instead of the C2 head.
(iii) C1 C2 C3
wh-u wh-u wh-u
Q-i EPP Q-u
EPP EPP
126
This combination of heads will not permit any derivation to converge. The wh-phrase
will be frozen in the specifier of C3 because all of its features will have been valued.
However, a wh-expletive (having only the [Q-u] feature) that originates in the clause
beneath the C2 head (with no [Q] feature) will not be attracted by that head because
they have no matching features. As predicted, no derivation will converge given this
selection. It is also possible that both the C2 and C3 heads could have a [Q-u] feature,
as in (iv).
(iv) C1 C2 C3
wh-u wh-u wh-u
Q-i Q-u Q-u
EPP EPP EPP
In this case, the wh-phrase originating in the lowest clause will be frozen in the
specifier of C3 because all of its features will have been valued. So wh-movement to
either the specifier of C1 or C2 will not be possible. However, the construction with a
wh-expletive in the specifier of C1 and C2 is possible, as no features will remain
unvalued. This is the only unexpected converging derivation.
It is worth considering whether this system of heads would permit constructions
in which there are only wh-expletives in every clause (no full wh-phrases). This
would not be possible because the [wh-u] feature on C3 head in (i)-(ii) would not be
valued, since the expletive has no [wh] feature. This derivation would fail to
converge. So too would a derivation in which full wh-phrases instead of wh-
expletives occupied the specifiers of C1, C2, and C3, though the intended
127
interpretation was the same as a canonical wh-expletive construction. Assuming the
heads were either like those in (i) or (ii) the [Q-u] on the wh-phrase in the specifier of
C3 would not be valued, because either the C2 head will not have its own [Q] feature
(as in (i)), or the C2 Probe will be inactive before probing the wh-phrase in the
specifier of C3, because it will have been valued through interaction with the wh-
phrase in its own clause. In either case, this type of derivation also will fail to
converge. The important thing to conclude from the exercise in this appendix is that
in this view the feature bundles that appear on the relevant heads permit all and only
the family of grammatical constructions, while making no use of stipulations or
devices particular to the construction under analysis here.